Fully automated detection pelvic lymph nodes in diffusion-weighted imaging for prostate cancer using deep learning: A multicenter study

Motivation: Accurate identification of lymph nodes (LNs) and assessment of metastatic burden are essential for effective clinical decision-making in prostate cancer treatment. Goal(s): To develop an LN segmentation model utilizing the V-Net on pelvic DWI and validate its performance across external datasets. Approach: We trained the 3D V-Net model on pelvic DWI images from 1,151 patients with 32,507 annotated LNs and validated it on data from 401 patients with 7,707 LNs across four hospitals. Results: The model achieved a sensitivity of 60.1% (95%CI, 57.6%-62.6%), a positive predictive value of 79.2% (95%CI, 76.6%-81.5%), and a false positive volume of 0.56 at the LN level. Impact: The results confirmed the feasibility of this method, which could aid in LN staging, quantitative measurements of tumor burden, and image-guided treatment of patients with PCa.